Multi-GNSS Constellation Anomaly Detection and Performance Monitoring

Abstract

This paper describes a process by which Global Navigation Satellite System (GNSS) nominal statistics and fault rates can be assessed with high confidence. The use of a GNSS in Advanced Receiver Autonomous Integrity Monitoring (ARAIM) requires knowledge of the level of performance of the signal-in-space (SIS) user range error (URE) in both nominal and faulty conditions. The performance characterization is done through a careful analysis of historical data, from which a determination of whether or not it meets the performance commitments of the constellation service provider (CSP) can be made. Typically, this analysis is carried out though a comparison of the broadcast navigation messages with post-processed precise estimates of the satellite clock and orbit states. This paper highlights the issues with such an approach as well as mitigation strategies to those issues. The issues primarily impact the estimates of fault rates rather than nominal statistics, as the number of faults is small compared to the overall amount of nominal data. In particular, methods of ensuring integrity in the logged broadcast navigation messages and precise clock and orbit products used in the analysis are described. In order to verify precise clock and orbit products, a Kalman filter (KF) to produce independent GNSS clock bias estimates has been developed and tested. The KF leverages the International GNSS Service (IGS) receiver network to produce estimates of GNSS clock biases given precisely known positions of the receivers as well as precise satellite orbit products. The KF serves multiple purposes in detecting and verifying faults. First, when only low rate (5 or 15 minute) clock products are available, higher rate products can be produced to detect short faults. Second, when precise clock products are available, the KF serves as a secondary check to protect against erroneous external precise products, which have been observed. Initial testing of the filter has shown an RMS error of under 20 centimeters for a five day run of GPS clock bias estimates, which is sufficient performance for precise product verification and fault detection.